JP5156452B2 - Defect classification method, program, computer storage medium, and defect classification apparatus - Google Patents

Description

  The present invention relates to a method, a program, a computer storage medium, and a defect classification device for classifying defects on a substrate based on a captured image of the substrate.

  For example, in a photolithography process in the manufacture of a semiconductor device, for example, a resist coating process in which a resist solution is applied on a semiconductor wafer (hereinafter referred to as “wafer”) to form a resist film, and a predetermined pattern is exposed on the resist film. An exposure process, a development process for developing the exposed resist film, and the like are sequentially performed to form a predetermined resist pattern on the wafer.

  For wafers that have undergone a series of photolithography processes, whether or not a predetermined resist film has been formed on the wafer surface by an inspection device, or whether or not appropriate exposure processing has been performed, as well as scratches and foreign matter A so-called macro defect inspection for inspecting whether or not there is adhesion is performed.

  In such a macro defect inspection, while moving the mounting table on which the wafer is mounted, the wafer on the mounting table is illuminated, and an image of the wafer is captured by, for example, an imaging device of a CCD line sensor. Image processing is performed to determine the presence or absence of defects (Patent Document 1).

  And when determining the presence or absence of a defect, conventionally classifying a defect using a method called learning type classification has been performed. In learning type classification, defect images can be collected in advance as teaching images, and defects can be classified optimally by learning them.

JP 2007-240519 A

  However, in order to classify defects optimally by conventional learning type classification, it is necessary to collect a large amount of defect images in advance, so that there is no such defect image at the start of the production process, and defects can be appropriately classified. could not. In addition, when defects are classified using only a small number of defect images, an overfitting phenomenon of learning with respect to defect images called overlearning occurs, and it is difficult to properly classify defects.

  The present invention has been made in view of such a point, and even when there is no defect image or when there are only a few defect images, the defect of the substrate is appropriately classified from the image to be inspected of the substrate. Objective.

  In order to achieve the above object, the present invention provides a defect classification method for classifying defects on a substrate based on an image to be inspected of the substrate, wherein the defect classification class is determined based on a feature amount of the defect. A design step of setting and storing a relationship between the feature amount of the defect and the classification class in a storage unit, and a feature amount calculation step of calculating a feature amount of the defect of the substrate from the imaged inspection target image of the substrate And a classification step of classifying the defect of the substrate into the classification class from the relationship between the characteristic quantity of the defect stored in the storage unit and the classification class based on the calculated feature quantity of the defect. The design step includes: a first step of creating a plurality of defect templates; a second step of generating a defect model by synthesizing the defect template with a teaching image of a substrate having no defect; and Defect model A third step of calculating a defect feature amount in the defect, a fourth step of setting a defect classification class for the defect feature amount in the defect model, and the defect feature amount and the classification class. And a fifth step of storing the relationship in the storage unit.

  According to the present invention, a defect template is created in advance, and a defect model is generated by synthesizing the defect template and a teaching image having no defect. Therefore, the defect model is used as a defect image in the conventional learning type classification method. Can be used. Then, a defect classification class can be set for the defect feature quantity in the defect model, and the relationship between the defect feature quantity and the defect classification class can be stored in the storage unit. Therefore, even when there is no defect image or there are only a few defect images, the defect feature amount is calculated from the image to be inspected of the imaged substrate, and the defect feature amount and the defect classification class stored in the storage unit Thus, the substrate defects can be classified appropriately.

  In the second step, the defect model may be generated by synthesizing a teaching image of the substrate having no defect and a plurality of the defect templates.

  In the design process, the feature amount and the classification class of the defect in the storage unit may be linked with unique information of the substrate. The specific information held by the substrate includes a substrate ID, a substrate lot ID, a substrate processing condition, a substrate processing date and time, and the like.

In the design process, when the defect is set to a different classification class for the same feature quantity of the defect, either one of the classification classes is deleted or both of the classification classes are deleted. May be.

  In the design process, when there is a teaching image of a defective substrate, a defect feature amount is calculated from the teaching image, and a defect classification class is set for the defect feature amount in the teaching image. The relationship between the feature quantity of the defect and the classification class may be further stored in the storage unit.

  The substrate may be divided into a plurality of inspection regions, the feature amount of the defect in each inspection region may be calculated in the feature amount calculation step, and the inspection region may be classified into the classification class in the classification step. .

  After the classification step, the classification class of the defect of the substrate is confirmed, and if it is determined that the classification class is misclassification, the relationship between the feature quantity of the defect in the storage unit and the classification class May be corrected.

  After the classification step, the substrate defect may be classified into the classification category based on a relationship between a plurality of predetermined classification classes and a single classification category.

  According to another aspect of the present invention, in order to cause the defect classification apparatus to execute the defect classification method, a program that operates on a computer of the defect classification apparatus is provided.

  According to another aspect of the present invention, a readable computer storage medium storing the program is provided.

  According to another aspect of the present invention, a defect classification apparatus for classifying defects on a substrate based on an image to be inspected of a substrate, wherein the defect classification class is set based on a feature amount of the defect. And a diagnostic means for classifying the defect of the substrate into the classification class set by the design means from the imaged inspection target image of the substrate, and the design means includes a plurality of defect templates. A stored template storage unit, a teaching image of a substrate having no defect, and the defect template are combined to generate a defect model; a defect feature amount in the defect model is calculated; A classification class setting unit that sets a defect classification class for the feature quantity; and a storage unit that stores a relationship between the feature quantity of the defect and the classification class. A feature amount calculation unit that calculates a feature amount of a defect of the substrate from the imaged inspection target image of the substrate, and a feature amount of the defect stored in the storage unit based on the calculated feature amount of the defect A classification unit that classifies the defects of the substrate into the classification class based on the relationship with the classification class.

  The model generation unit may generate a defect model by combining a teaching image of a substrate having no defect and the plurality of defect templates.

The design unit may have a storage unit history management function that stores the feature amount of the defect and the classification class in the storage unit in a linked manner with unique information of the substrate. The specific information held by the substrate includes a substrate ID, a substrate lot ID, a substrate processing condition, a substrate processing date and time, and the like.

In the storage unit, when the defect is set to a different classification class for the same feature quantity of the defect , the design unit deletes one of the classification classes or both You may have the memory | storage part check function which deletes a classification class.

  The design unit includes another classification class setting unit that calculates a defect feature amount from a teaching image of a defective substrate and sets a defect classification class for the defect feature amount. A relationship between the feature quantity and the classification class may be further stored in the storage unit.

  The diagnostic unit includes a preprocessing unit that divides the substrate into a plurality of inspection regions, the feature amount calculation unit calculates feature amounts of defects in the inspection regions, and the classification unit includes the inspection units. The region may be classified into the classification class.

  The diagnostic means includes a confirmation unit that confirms the classification class of the defect of the substrate classified by the classification unit, and when the confirmation unit determines that the classification class is misclassification, the storage unit The relationship between the feature quantity of the defect in the part and the classification class may be corrected.

  The diagnostic means includes a post-processing unit that associates a plurality of classification classes with a single classification category, and the post-processing unit identifies defects on the substrate based on the association with a plurality of classification categories. May be classified.

  According to the present invention, even when there is no defect image or when there are only a few defect images, the defect of the substrate can be appropriately classified from the inspection target image of the imaged substrate.

  Hereinafter, preferred embodiments of the present invention will be described. FIG. 1 is a plan view schematically showing the configuration of a coating and developing treatment system 1 equipped with the defect classification apparatus according to the present embodiment, FIG. 2 is a front view of the coating and developing treatment system 1, and FIG. 2 is a rear view of the coating and developing treatment system 1. FIG.

  As shown in FIG. 1, the coating and developing treatment system 1 is a cassette that carries, for example, 25 wafers W from the outside to the coating and developing treatment system 1 in a cassette unit, and carries a wafer W into and out of the cassette C. A station 2, a processing station 3 in which a plurality of various processing apparatuses for performing predetermined processing in a single wafer type in a photolithography process are arranged in multiple stages, and an exposure apparatus provided adjacent to the processing station 3 It has a configuration in which an interface station 4 that transfers the wafer W to and from (not shown) is integrally connected.

  The cassette station 2 is provided with a cassette mounting table 5 that can mount a plurality of cassettes C in a row in the X direction (vertical direction in FIG. 1). The cassette station 2 is provided with a wafer transfer body 7 that can move in the X direction on the transfer path 6. The wafer carrier 7 is also movable in the wafer arrangement direction (Z direction; vertical direction) of the wafers W accommodated in the cassette C, and is selective to the wafers W in each cassette C arranged in the X direction. Is accessible.

  The wafer transfer body 7 is rotatable in the θ direction around the Z axis, and a temperature control device 60 belonging to a third processing device group G3 on the processing station 3 side to be described later, and a transition device 61 for delivering the wafer W. Is also accessible.

  The processing station 3 adjacent to the cassette station 2 includes, for example, five processing device groups G1 to G5 in which a plurality of processing devices are arranged in multiple stages. A first processing device group G1 and a second processing device group G2 are arranged in this order from the cassette station 2 side on the X direction negative direction (downward direction in FIG. 1) side of the processing station 3. A third processing device group G3, a fourth processing device group G4, and a fifth processing device group G5 are sequentially arranged from the cassette station 2 side on the X direction positive direction (upward direction in FIG. 1) side of the processing station 3. Is arranged. A first transfer device A1 is provided between the third processing device group G3 and the fourth processing device group G4, and the wafer W is supported and transferred inside the first transfer device A1. A first transfer arm 10 is provided. The first transfer arm 10 can selectively access each processing unit in the first processing unit group G1, the third processing unit group G3, and the fourth processing unit group G4 to transfer the wafer W. it can. A second transfer device A2 is provided between the fourth processing device group G4 and the fifth processing device group G5, and the wafer W is supported and transferred inside the second transfer device A2. A second transfer arm 11 is provided. The second transfer arm 11 can selectively access each processing apparatus in the second processing apparatus group G2, the fourth processing apparatus group G4, and the fifth processing apparatus group G5 to transfer the wafer W. it can.

  As shown in FIG. 2, in the first processing unit group G1, a liquid processing apparatus that performs processing by supplying a predetermined liquid to the wafer W, for example, resist coating apparatuses 20, 21, and 22 that apply a resist solution to the wafer W. Bottom coating devices 23 and 24 for forming an antireflection film for preventing reflection of light during the exposure process are stacked in five stages in order from the bottom. In the second processing unit group G2, liquid processing units, for example, development processing units 30 to 34 for supplying a developing solution to the wafer W and performing development processing are stacked in five stages in order from the bottom. In addition, chemical chambers 40 and 41 for supplying various processing liquids to the liquid processing apparatuses in the processing apparatus groups G1 and G2 are provided at the bottom of the first processing apparatus group G1 and the second processing apparatus group G2. Each is provided.

  As shown in FIG. 3, the third processing unit group G3 includes a temperature control unit 60, a transition unit 61, high-precision temperature control units 62 to 64 that control the temperature of the wafer W under high-precision temperature control, and a defect inspection unit. 110 and 110 are stacked in seven steps from the bottom. As shown in FIG. 1, the defect inspection apparatus 110 is connected with a defect classification apparatus 200 that classifies defects on the wafer W from the inspection target image captured by the defect inspection apparatus 110.

  As shown in FIG. 3, the fourth processing unit group G4 includes, for example, a high-precision temperature control unit 70, pre-baking units 71 to 74 that heat-treat the wafer W after the resist coating process, and a wafer W that has been developed. Post baking apparatuses 75 to 79 to be processed are stacked in 10 stages in order from the bottom.

  In the fifth processing unit group G5, a plurality of heat processing devices for heat-treating the wafer W, for example, high-precision temperature control devices 80 to 83, and post-exposure baking devices 84 to 89 for heat-processing the exposed wafer W are sequentially arranged from the bottom. It is stacked in 10 steps.

  As shown in FIG. 1, a plurality of processing devices are arranged on the positive side in the X direction of the first transfer device A1, and for example, an adhesion device 90 for hydrophobizing the wafer W as shown in FIG. 91, and heating devices 92 and 93 for heating the wafer W are stacked in four stages in order from the bottom. As shown in FIG. 1, a peripheral exposure device 94 that selectively exposes only the edge portion of the wafer W, for example, is disposed on the positive side in the X direction of the second transfer device A2.

  In the interface station 4, for example, as shown in FIG. 1, a wafer transfer body 101 moving on a transfer path 100 extending in the X direction and a buffer cassette 102 are provided. The wafer transfer body 101 can move in the Z direction and can also rotate in the θ direction. With respect to the exposure apparatus (not shown) adjacent to the interface station 4, the buffer cassette 102, and the fifth processing apparatus group G5. The wafer W can be transferred by accessing.

  Next, configurations of the defect inspection apparatus 110 and the defect classification apparatus 200 described above will be described.

  The defect inspection apparatus 110 has a casing 111 as shown in FIG. A loading / unloading port 112 for loading / unloading the wafer W is formed on one end side of the casing 111 (X direction negative direction side in FIG. 4) on both side faces of the casing 111 facing the short direction. At the loading / unloading port 112, an opening / closing shutter 113 is provided.

  In the casing 111, a mounting table 120 on which the wafer W is mounted is provided as shown in FIG. The mounting table 120 can be freely rotated and stopped by a rotation drive unit 121 such as a motor, and has an alignment function for adjusting the position of the wafer W. A guide rail 122 extending from one end side (X direction negative direction side in FIG. 5) to the other end side (X direction positive direction side in FIG. 5) is provided on the bottom surface of the casing 111. . The mounting table 120 and the rotation driving unit 121 are provided on the guide rail 122 and can be moved along the guide rail 122 by a driving device 123 such as a pulse motor.

  A buffer arm 124 that temporarily supports the wafer W is provided at a position P1 (a position indicated by a solid line in FIG. 5) on one end side in the casing 111 where the wafer W is carried into and out of the casing 111. . As shown in FIG. 6, the buffer arm 124 has a support portion 124a for the wafer W at the tip. The support portion 124a is formed in, for example, a 3/4 annular shape. The diameter of the 3/4 circular ring of the support part 124a is larger than the diameter of the mounting table 120, and the mounting table 120 can be accommodated inside the support part 124a. The 3/4 circular notch portion of the support portion 124a is formed on the other end side in the casing 111 (the X direction positive direction side in FIG. 6), and the mounting table 120 interferes with the support portion 124a. Without moving to the other end. A plurality of support pins 124b are provided on the support portion 124a, and the wafer W is supported on the support pins 124b. The base 124 c of the buffer arm 124 is attached to an elevating drive unit 125 such as a cylinder, and the buffer arm 124 can be moved up and down the mounting table 120.

  As shown in FIG. 5, on the other end side of the casing 111, the alignment position P2 (position indicated by a dotted line in FIG. 5) for adjusting the position of the notch portion of the wafer W is the position of the wafer W on the mounting table 120. A sensor 126 for detecting the position of the notch portion is provided. The position of the notch portion of the wafer W can be adjusted by rotating the mounting table 120 by the rotation driving unit 121 while detecting the position of the notch portion by the sensor 126.

  An imaging device 130 is provided on a side surface on the other end side in the casing 111 (X direction positive direction side in FIG. 5). For the imaging device 130, for example, a wide-angle CCD camera is used. Near the upper center of the casing 111, a half mirror 131 is provided. The half mirror 131 is provided at a position facing the imaging device 130 and is inclined by 45 degrees from the vertical direction. An illumination device 132 that can change the illuminance is provided above the half mirror 131, and the half mirror 131 and the illumination device 132 are fixed to the upper surface of the casing 111. Further, the imaging device 130, the half mirror 131, and the illumination device 132 are respectively provided above the wafer W placed on the mounting table 120. The illumination from the illumination device 132 passes through the half mirror 131 and is illuminated downward. Therefore, the reflected light of the object in the irradiation area is reflected by the half mirror 131 and is taken into the imaging device 130. That is, the imaging device 130 can image an object in the irradiation area. The captured inspection target image of the wafer W is output to the defect classification apparatus 200.

  As illustrated in FIG. 7, the defect classification apparatus 200 includes a design unit 201 that sets a defect classification class based on a defect feature amount in advance, and an inspection target image of the wafer W that is captured by the defect inspection apparatus 110. And a diagnostic unit 202 that classifies the defects into the classification class set by the design unit 201.

  The design unit 201 includes a teaching image input unit 210 to which a teaching image captured by the defect inspection apparatus 110 is input before an inspection target image is captured. As the teaching image, either an image of the wafer W having no defect or an image of the wafer W having a defect is input.

  The teaching image input to the teaching image input unit 210 is output to the pre-design processing unit 211. In the pre-design processing unit 211, the teaching image of the wafer W is divided into chips C as inspection regions as shown in FIG. The inspection area is not limited to the area of the chip C, and can be arbitrarily set by the user.

  As shown in FIG. 7, the design unit 201 includes a template storage unit 212 that stores a template imitating a defect of the wafer W (hereinafter referred to as “defect template”). As defects of the defect template, for example, defects such as scratches (scratches), particles, hot spots, and defocus are imitated by the defect template. A defect template is created for each chip C. For example, as shown in FIG. 9, a defect template T having a substantially circular defect D (FIG. 9A), a defect template T having an elongated elliptical defect D (FIG. 9). 9 (b)), a defect template T having a defect D on the entire surface of the chip C (FIG. 9C), and a defect template T having a polygonal defect D (FIG. 9D) are stored in the template storage unit 212. ing. These defects D are assigned defect numbers, and the defect templates T are assigned identification codes, which are managed in the template storage unit 212.

  As the defect template T, as shown in FIG. 10, only the defect M is extracted from the defect image of the chip C on the wafer W acquired in a separate process from the wafer W to be inspected and the image of the chip C without defect. The image of the defect D may be binarized and stored in the template storage unit 212. The defect template T can be arbitrarily added by the user.

  When the teaching image input to the teaching image input unit 210 is the image N1 of the wafer W having no defect, the teaching image N1 divided by the pre-design processing unit 211 is output to the model generation unit 213. The As shown in FIG. 11A, the model generation unit 213 can generate a defect model M by combining the teaching image N1 and the defect template T stored in the template storage unit 212. The defect model M is synthesized by arbitrarily selecting the brightness of the defect template T, the size of the defect D, the position of the defect D, the angle of the defect D, the number of the defects D, and the like. The defect model M may be generated by combining the teaching image N1 and a plurality of defect templates T as shown in FIG. The model generation unit 213 may be provided with a software GUI (Graphical User Interface) for assisting generation of the defect model M.

  As shown in FIG. 7, the defect model M generated by the model generation unit 213 is output to the first classification class setting unit 214 that sets the classification class of the defect D in the defect model M. In the first classification class setting unit 214, first, the feature amount of the defect D in the defect model M is calculated. The feature amount of the defect D includes, for example, the features of the defect D, such as light and shade, color, spatial features such as texture, or geometric features such as shape features (the size, shape, length, and width of the defect D). Calculated. Then, the classification class of the defect D is set for the calculated feature amount of the defect D.

  If the teaching image input to the teaching image input unit 210 is the image N2 of the defective wafer W, the teaching image N2 divided by the pre-design processing unit 211 has the second classification class. The data is output to the setting unit 215. The second classification class setting unit 215 calculates the feature quantity of the defect D in the teaching image N2, and sets the classification class of the defect D for the calculated feature quantity of the defect D.

  The feature amount and the classification class of the defect D set by the first classification class setting unit 214 and the second classification class 215 are output to the storage unit 220.

  The feature amount and classification class of the defect D output to the storage unit 220 are stored by being linked with unique information (tag 1 information described later) of the wafer W as shown in FIG. 12 by the storage unit history management function 221. Stored in the unit 220. The storage unit 220 manages and stores each defect classification class (Class in FIG. 12). Each classification class stores a defect D number (Data in FIG. 12), Tag1 information, Tag2 information, and feature data of the defect D in a linked manner. Tag1 information includes the ID of the wafer W, the ID of the lot of the wafer W, the ID of the device on which the wafer W was processed, the ID of the layer on the wafer W, the ID of the slot of the wafer W, the ID of the chip C, the ID of the wafer W The position information of the chip C, the processing recipe information of the wafer W, the processing date and time of the wafer W, and the like are stored. As Tag2 information, whether or not the defect template T is used when the model generation unit 213 generates the defect model M (for example, “0” is stored if used, and “1” is stored if not used). The identification code of the defect template T, the synthesis conditions (the brightness of the defect template T, the angle of the defect D, the position of the defect D, etc.) are stored. Thus, by linking and storing the accompanying information to the classification class of each defect D, the history of the accompanying data can be easily retrieved and extracted, and the learning history stored in the storage unit 220 is stored. Can be managed. When a sufficient data storage area can be secured in the storage unit 220, the defect model M and the teaching images N1 and N2 can be stored. Further, the design unit 201 may be provided with a plurality of storage units 220.

  Further, in the storage unit 220, it is possible to check whether or not different classification classes are stored for the feature quantity of the same defect D by the storage unit check function 222 shown in FIG. Then, when different classification classes are stored for the feature quantity of the same defect D as shown in FIG. 13, the user is notified of the presence or absence and either one of the classification classes is deleted, or The user can select whether to delete both classification classes. With this function, one classification class can be stored in the storage unit 220 for the feature amount of one defect D, and the defect D can be appropriately classified.

  As illustrated in FIG. 7, the diagnosis unit 202 includes an inspection target image input unit 230 into which an inspection target image of the wafer W imaged by the defect inspection apparatus 110 is input.

  The inspection target image input to the inspection target image input unit 230 is output to the preprocessing unit 231. In the pre-processing unit 231, the inspection target image E of the wafer W is divided for each chip C as an inspection region, similarly to the division processing in the design pre-processing unit 211 shown in FIG.

  The inspection target image E divided by the preprocessing unit 231 is output to the feature amount calculation unit 232 as shown in FIG. The feature amount calculation unit 232 calculates the feature amount of the defect D for each chip C of the inspection target image E. As the feature quantity of the defect D, the same parameters as the feature quantity stored in the storage unit 220 are calculated. For example, the density of the defect D, features such as color, spatial features such as texture, or geometric features such as shape features. Features are calculated.

  The feature amount of the defect D calculated by the feature amount calculation unit 232 is output to the classification unit 233 that classifies the defect D into a classification class. The classification unit 233 classifies the defects into classification classes using a learning type classification method based on the relationship between the feature amount of the defect D stored in the storage unit 220 and the classification class. As the learning type classification method, for example, a neural network, a k-NN (k-Nearest Neighbor), a TFC (Test Feature Classifier), or the like is used. Note that a learning / training unit 223 is provided in the design unit 201 in order to make the data in the storage unit 220 correspond to such a learning classification method. In the learning / training unit 223, for example, in consideration of learning efficiency (calculation time and classification system), “selection / extraction” of feature quantity data in the storage unit 220, “optimization of weight coefficient”, “unusable data” Delete "is performed. Further, when there are a plurality of storage units 220, switching and selection are performed.

  The diagnostic unit 202 includes a post-processing unit 234 that associates a plurality of classification classes with a single classification category. In the post-processing unit 234, for example, when there is a problem that the classification class classified by the classification unit 233 is too subdivided, the defect D is classified into the classification category based on the association between the classification class and the classification category described above. The For example, in the case of the classification classes Class 3 to 6 of the defect D, the defects are classified into the classification category Category1.

  The diagnosis unit 202 includes a confirmation unit 235 that allows the user to check the suitability of the classification class classified by the classification unit 233 or the classification category classified by the post-processing unit 234 in a timely manner. In the confirmation unit 235, when the user determines that the classification class or classification category of the defect D is misclassification, the feature amount of the defect D is output to the classification class correction unit 224 of the design unit 201. Then, a correct classification class is set by the classification class correction unit 224, the result is output to the storage unit 220, and the classification class in the storage unit 220 is corrected.

  The diagnosis unit 202 has a report unit 236 that visualizes and reports the classified defects D on the wafer W. The reporting unit 236 displays the classification class of the defect D for each chip C on the wafer W as shown in FIG. 14A, and also displays the classification class of the defect D as shown in FIG. b) a correspondence table between the classification result) and the feature amount is displayed.

  The defect classification apparatus 200 including the design unit 201 and the diagnosis unit 202 is, for example, a computer, and has a program for executing the above-described defect classification of the wafer W. The program is recorded on a computer-readable storage medium such as a hard disk (HD), a compact disk (CD), a magnetic optical disk (MO), or a memory card, and defect classification is performed from the storage medium. It may be installed in the apparatus 200.

  Next, the wafer processing performed in the entire coating and developing treatment system 1 for the inspection of the defects of the wafer W performed by the defect inspection apparatus 110 configured as described above and the classification of the defects of the wafer W performed by the defect classification apparatus 200. This will be explained together with the process.

  First, one wafer W is taken out from the cassette C on the cassette mounting table 5 by the wafer transfer body 7 and transferred to the temperature adjustment device 60 of the third processing unit group G3. The wafer W transferred to the temperature adjusting device 60 is adjusted to a predetermined temperature, and then transferred to the bottom coating device 23 by the first transfer arm 10 to form an antireflection film. The wafer W on which the antireflection film is formed is sequentially transferred to the heating device 92 and the high-precision temperature control device 70 by the first transfer arm 10 and subjected to predetermined processing in each device. Thereafter, the wafer W is transferred to the resist coating apparatus 20.

  When a resist film is formed on the wafer W in the resist coating apparatus 20, the wafer W is transferred to the pre-baking apparatus 71 by the first transfer arm 10, and then the peripheral exposure apparatus 94 and the high exposure apparatus 94 are moved by the second transfer arm 11. It is sequentially conveyed to the precision temperature control device 83, and predetermined processing is performed in each device. Thereafter, the wafer is transferred to an exposure apparatus (not shown) by the wafer transfer body 101 of the interface station 4, and a predetermined pattern is exposed on the resist film on the wafer W. The wafer W that has been subjected to the exposure process is transferred to the post-exposure baking apparatus 84 by the wafer transfer body 101 and subjected to a predetermined process.

  When the heat treatment in the post-exposure baking apparatus 84 is completed, the wafer W is transferred to the high-accuracy temperature adjustment apparatus 81 by the second transfer arm 11 and the temperature is adjusted, and then transferred to the development processing apparatus 30 and developed on the wafer W. Is applied to form a pattern on the resist film. Thereafter, the wafer W is transferred to the post-baking device 75 by the second transfer arm 11 and subjected to heat treatment, and then transferred to the high-accuracy temperature adjusting device 63 to adjust the temperature. Then, the wafer W is transferred to the defect inspection apparatus 110 by the first transfer arm 10, and defect inspection of the wafer W is performed. Details of this defect inspection will be described later. Thereafter, the wafer W is transferred to the transition device 61 by the first transfer arm 10 and returned to the cassette C by the wafer transfer body 7 to complete a series of photolithography steps.

  Next, a defect inspection method for the wafer W in the defect inspection apparatus 110 and a defect classification method for the wafer W in the defect classification apparatus 200 will be described.

  First, the wafer W transferred into the casing 111 by the first transfer arm 10 is mounted on the mounting table 120. Then, the illumination device 132 illuminates the wafer W with a predetermined illuminance when the wafer W passes under the half mirror 131 while moving the mounting table 120 toward the imaging device 130. In this way, an image of the wafer W is taken by the imaging device 130 while illuminating the wafer W. The captured inspection image E of the wafer W is output to the inspection target image input unit 230 of the defect classification apparatus 200. The wafer W on which the inspection object image E has been taken is moved to the loading / unloading port 112 side, then transferred from the buffer arm 124 to the wafer carrier 7 and unloaded from the defect inspection apparatus 110 by the wafer carrier 7. The

  The inspection target image E of the wafer W input to the inspection target image input unit 230 is output to the preprocessing unit 231 and divided for each chip C. Then, the divided inspection target image E is output to the feature amount calculation unit 232, and the feature amount of the defect D is calculated from the inspection target image E.

  The feature amount of the defect D of the inspection target image E calculated by the feature amount calculation unit 232 is output to the classification unit 233. The classification unit 233 classifies defects using a learning type classification method. In the present embodiment, a case will be described in which a k-NN (k-Nearest Neighbor) method is used as a learning type classification method. This k-NN method compares the feature amount of the defect D input from the feature amount calculation unit 232 with the feature amount data of the defect D stored in the storage unit 220 in advance in the feature amount space. The most similar (closest distance) feature quantity data, that is, the classification class to which the nearest neighbor belongs. That is, by increasing the definition (k value) of the nearest neighbor, a more suitable classification class is output from the frequency distribution of the classification classes to which the top k pieces of similar feature amount data belong.

  The classification method in the classification unit 233 will be specifically described. First, the deviation sum of squares (distance) between the feature amount of the defect D from the feature amount calculation unit 232 and all the feature amount data in the storage unit 220 is calculated. Next, the calculated distance data is rearranged in ascending order. The top k pieces of the sorted distance data are extracted. Find the classification class with the highest frequency in the frequency distribution of the extracted data. Then, the classification class is output as the classification result of the defect D.

  In this way, when the classification class classified by the classification unit 233 is a classification class that can be used as it is in all the chips C of the wafer W, the classification class is output to the reporting unit 236. Then, in the reporting unit 236, the classification class of the defect D for each chip C on the wafer W is displayed as shown in FIG. 14A, and the classification class of the defect D as shown in FIG. A correspondence table with the feature values is displayed.

  On the other hand, when there is a problem that the classification class classified by the classification unit 233 is, for example, too subdivided, the post-processing unit 234 determines that the defect D is classified based on the above-described association between the classification class and the classification category. are categorized. For example, in the case of the classification classes Class 3 to 6 of the defect D, the defect D is classified into the classification category Category1. Then, the classification category is output to the report unit 236, and the classification category for each chip C is visualized on the wafer W.

  Note that the user can check the suitability of the classification class classified by the classification unit 233 or the classification category classified by the post-processing unit 234 in a timely manner in the confirmation unit 235. In the confirmation unit 235, when the user determines that the classification class or classification category of the defect D is misclassification, the feature amount of the defect D is output to the classification class correction unit 224 of the design unit 201. A correct classification class is set by the classification class correction unit 224, and the result is output to the storage unit 220. In this way, the classification class of the storage unit 220 is corrected.

  According to the above embodiment, the defect template T is created in advance and stored in the template storage unit 212, and the model generation unit 213 generates the defect model M by synthesizing the defect template T and the teaching image N1 having no defect. Therefore, the defect model M can be used as a defect image in the conventional learning type classification method. Then, in the first classification class setting unit 214, the feature amount of the defect D in the defect model M is calculated, the classification class of the defect D is set for the feature amount of the defect D, and the feature amount of the defect D and the defect The relationship with the classification class of D can be stored in the storage unit 220. Therefore, even when there is no defect image or there are only a few defect images, the feature amount calculation unit 232 calculates the relationship between the feature amount of the defect D stored in the storage unit 220 and the classification class of the defect D. The defect D of the wafer W can be appropriately classified from the feature amount of the defect D of the wafer W.

  In addition, since the image input to the teaching image input unit 210 is divided into chips C as inspection areas in the design preprocessing unit 211, the defect D of the wafer W can be further subdivided and appropriately classified. .

  Further, in the storage unit 220, the storage unit history management function 221 links the feature amount and classification class of the defect D with unique information (such as the Tag1 information described above) that the wafer W has. Can be easily retrieved and extracted, and the learning history stored in the storage unit 220 can be managed.

  In the storage unit 220, the storage unit check function 222 can check whether or not different classification classes are stored for the feature amount of the same defect D. If the defect D is set to a different classification class for the feature quantity of the same defect D, either one of the classification classes can be deleted, or both classification classes can be deleted. Therefore, one classification class can be stored in the storage unit 220 for the feature amount of one defect D, and the defect D can be appropriately classified.

  When the teaching image input to the teaching image input unit 210 is the image N1 of the defective wafer W, the second classification class setting unit 215 determines the feature amount of the defect D in the teaching image N1. Since the classification class of the defect D is set for the calculated feature quantity of the defect D, the relationship between the feature quantity of the defect D and the classification class can be further stored in the storage unit 220. As a result, the relationship between the feature quantity of the defect D stored in the storage unit 220 and the classification class can be increased, and since the actual defect data is input, the accuracy can be improved.

  In addition, since the post-processing unit 234 associates a plurality of classification classes with a single classification category, for example, even if the classification class classified by the classification unit 233 has a problem such as over-division, D can be classified into an appropriate classification category.

  Further, in the confirmation unit 235, since the user can confirm in a timely manner whether or not the classification class classified by the classification unit 233 or the classification category classified by the post-processing unit 234 is correct, the erroneous classification of the defect D can be prevented. it can. Further, when it is determined that the classification class or the classification category is misclassification, the classification class correction unit 224 corrects the classification class for the feature amount of the defect D, so that the accuracy of the classification class in the storage unit 220 is increased. It can be improved further.

  In the above embodiment, the teaching images N1 and N2 input to the teaching image input unit 210 and the inspection target image E input to the inspection target image input unit 230 are each divided into chips C. The defect D of the entire wafer W may be classified without dividing the teaching images N1 and N2 and the inspection target image E. In such a case, as shown in FIG. 15, the division processing in the design preprocessing unit 211 and the preprocessing unit 231 can be omitted. As the defect template T stored in the template storage unit 212, for example, as shown in FIG. 16, the defect template T (FIG. 16 (a)) having a defect in the upper half of the wafer W, the upper right quarter of the wafer W. A defect template T having a defect in the wafer W (FIG. 16B), a defect template T having a ring-like defect in the peripheral portion of the wafer W (FIG. 16C), and an equal interval in the peripheral portion of the upper half of the wafer W A defect template T having a plurality of defects (FIG. 16D) or the like is used. In this case, the defect D of the entire wafer W can be classified by the same method as the defect classification method described above.

  The preferred embodiments of the present invention have been described above with reference to the accompanying drawings, but the present invention is not limited to such examples. It is obvious for those skilled in the art that various modifications or modifications can be conceived within the scope of the idea described in the claims, and these naturally belong to the technical scope of the present invention. It is understood. The present invention is not limited to this example and can take various forms. The present invention can also be applied to a case where the substrate is another substrate such as an FPD (flat panel display) other than a wafer or a mask reticle for a photomask.

  The present invention is useful when classifying defects on a substrate based on the image of the imaged substrate.

It is a top view which shows the outline of a structure of the coating and developing processing system carrying the defect classification device concerning this Embodiment. It is a front view of a coating and developing treatment system. It is a rear view of a coating and developing treatment system. It is a cross-sectional view which shows the outline of a structure of a defect inspection apparatus. It is a longitudinal cross-sectional view which shows the outline of a structure of a defect inspection apparatus. It is a top view which shows the outline of a structure of a buffer arm. It is a schematic diagram which shows the outline of a structure of a defect classification device. It is explanatory drawing which showed the test | inspection area | region on a wafer. It is the top view which illustrated the defect template. It is explanatory drawing which showed a mode that a defect template was produced | generated. It is explanatory drawing which showed a mode that a defect model was produced | generated. It is explanatory drawing which showed the data memorize | stored in the memory | storage part. It is explanatory drawing which shows the case where a different classification | category class is set with respect to the feature-value of the same defect in a memory | storage part. It is explanatory drawing which showed the classification | category of the defect reported by a report part. It is a schematic diagram which shows the outline of a structure of the defect classification | category apparatus in other embodiment. It is the top view which illustrated the defect template of the whole wafer.

Explanation of symbols

DESCRIPTION OF SYMBOLS 1 Application | coating development processing system 110 Defect inspection apparatus 200 Defect classification apparatus 201 Design means 202 Diagnosis means 210 Teaching image input part 211 Design pre-processing part 212 Template memory | storage part 213 Model generation part 214 1st classification class setting part 215 2nd Classification class setting unit 220 Storage unit 221 Storage unit history management function 222 Storage unit check function 223 Learning / training unit 224 Classification class correction unit 230 Inspection target image input unit 231 Preprocessing unit 232 Feature amount calculation unit 233 Classification unit 234 Postprocessing unit 235 Confirmation unit 236 Report unit C Chip D Defect E Image to be inspected M Defect model N1 Teaching image without defect N2 Teaching image without defect T Defect template W Wafer

Claims (18)

A defect classification method for classifying defects on a substrate based on an image to be inspected of a substrate,
Setting a defect classification class based on the feature quantity of the defect, and storing the relationship between the feature quantity of the defect and the classification class in a storage unit;
A feature amount calculation step of calculating a feature amount of a defect of the substrate from the imaged inspection target image of the substrate;
A classification step of classifying the defect of the substrate into the classification class based on the calculated feature quantity of the defect based on the relationship between the classification quantity and the feature quantity of the defect stored in the storage unit; ,
The design process includes
A first step of creating a plurality of defect templates;
A second step of generating a defect model by synthesizing the defect template with a teaching image of a substrate having no defect;
A third step of calculating a feature amount of the defect in the defect model;
A fourth step of setting a defect classification class for the defect feature quantity in the defect model;
A defect classification method comprising: a fifth step of storing a relationship between the feature quantity of the defect and the classification class in the storage unit. 2. The defect classification method according to claim 1, wherein, in the second step, the defect model is generated by synthesizing a teaching image of the substrate having no defect and a plurality of the defect templates. 3. 3. The defect classification method according to claim 1, wherein in the design step, the feature quantity and the classification class of the defect in the storage unit are linked with unique information of the substrate. In the design process, when the defect is set to a different classification class for the same feature amount of the defect, either one of the classification classes is deleted or both of the classification classes are deleted. The defect classification method according to claim 1, wherein: In the design process, if there is a teaching image of a defective substrate,
Calculate the feature amount of the defect from the teaching image,
A defect classification class is set for the feature amount of the defect in the teaching image,
The defect classification method according to claim 1, wherein a relationship between the feature quantity of the defect and the classification class is further stored in the storage unit. The substrate is divided into a plurality of inspection areas;
In the feature amount calculating step, the feature amount of the defect in each inspection area is calculated,
The defect classification method according to claim 1, wherein in the classification step, the inspection areas are classified into the classification classes. After the classification step, the classification class of the defect of the substrate is confirmed, and if it is determined that the classification class is misclassification, the relationship between the feature quantity of the defect in the storage unit and the classification class The defect classification method according to claim 1, wherein the defect classification method is corrected. The defect of the said board | substrate is classified into the said classification category based on the relationship between a predetermined several classification class and a single classification category after the said classification | category process of Claims 1-7 characterized by the above-mentioned. The defect classification method according to any one of the above. A program that operates on a computer of the defect classification apparatus in order to cause the defect classification apparatus to execute the defect classification method according to claim 1. A readable computer storage medium storing the program according to claim 9. A defect classification device that classifies defects on a substrate based on an image to be inspected of the substrate,
A design means for setting a defect classification class based on the feature amount of the defect;
A diagnostic unit that classifies the defect of the substrate into the classification class set by the design unit from the image to be inspected of the substrate,
The design means includes
A template storage unit in which a plurality of defect templates are stored;
A model generation unit that generates a defect model by synthesizing the defect template with a teaching image of a substrate having no defect,
A classification class setting unit that calculates a feature quantity of a defect in the defect model and sets a classification class of the defect with respect to the feature quantity of the defect;
A storage unit that stores a relationship between the feature quantity of the defect and the classification class;
The diagnostic means includes
A feature amount calculation unit that calculates a feature amount of a defect of the substrate from the imaged inspection target image of the substrate;
A classifying unit that classifies the defects on the substrate into the classification class based on the relationship between the feature quantity of the defect stored in the storage unit and the classification class based on the calculated feature quantity of the defect; A defect classification apparatus characterized by. The defect classification apparatus according to claim 11, wherein the model generation unit generates a defect model by combining a teaching image of a substrate having no defect and a plurality of the defect templates. The design unit has a storage unit history management function for storing the feature quantity and the classification class of the defect in the storage unit in a linked manner with unique information of the substrate. 12. The defect classification apparatus according to 12. In the storage unit, when the defect is set to a different classification class for the same feature quantity of the defect , the design unit deletes one of the classification classes or both The defect classification apparatus according to claim 11, further comprising a storage unit check function for deleting a classification class. The design means includes another classification class setting unit that calculates a feature amount of a defect from a teaching image of a substrate having a defect, and sets a classification class of the defect with respect to the feature amount of the defect,
The defect classification apparatus according to any one of claims 11 to 14, wherein a relationship between the feature quantity of the defect and the classification class is further stored in the storage unit. In the diagnostic means, a pre-processing unit that divides the substrate into a plurality of inspection regions,
The feature amount calculation unit calculates a feature amount of a defect in each inspection region,
The defect classification apparatus according to claim 11, wherein the classification unit classifies each inspection region into the classification class. The diagnostic means has a confirmation unit for confirming the classification class of the defect of the substrate classified by the classification unit,
The relationship between the feature quantity of the defect in the storage unit and the classification class is corrected when the classification unit determines that the classification class is a misclassification. The defect classification apparatus according to any one of 16. The diagnostic means includes a post-processing unit that associates a plurality of classification classes with a single classification category,
The defect classification apparatus according to claim 11, wherein the post-processing unit classifies the defect of the substrate into a plurality of classification categories based on the association.

Images